with_advisory_lock

Adds advisory locking (mutexes) to ActiveRecord 3.2, 4.0 and 4.1 when used with MySQL or PostgreSQL. SQLite resorts to file locking.

What's an "Advisory Lock"?

An advisory lock is a mutex used to ensure no two processes run some process at the same time. When the advisory lock is powered by your database server, as long as it isn't SQLite, your mutex spans hosts.

Usage

Where User is an ActiveRecord model, and lock_name is some string:

User.with_advisory_lock(lock_name) do
  do_something_that_needs_locking
end

What happens

1. The thread will wait indefinitely until the lock is acquired.
2. While inside the block, you will exclusively own the advisory lock.
3. The lock will be released after your block ends, even if an exception is raised in the block.

Lock wait timeouts

The second parameter for with_advisory_lock is timeout_seconds, and defaults to nil, which means wait indefinitely for the lock.

A value of zero will try the lock only once. If the lock is acquired, the block will be yielded to. If the lock is currently being held, the block will not be called.

Note that if a non-nil value is provided for timeout_seconds, the block will not be invoked if the lock cannot be acquired within that time-frame.

Return values

The return value of with_advisory_lock_result is a WithAdvisoryLock::Result instance, which has a lock_was_acquired? method and a result accessor method, which is the returned value of the given block. If your block may validly return false, you should use this method.

The return value of with_advisory_lock will be the result of the yielded block, if the lock was able to be acquired and the block yielded, or false, if you provided a timeout_seconds value and the lock was not able to be acquired in time.

Testing for the current lock status

If you needed to check if the advisory lock is currently being held, you can call Tag.advisory_lock_exists?("foo"), but realize the lock can be acquired between the time you test for the lock, and the time you try to acquire the lock.

If you want to see if the current Thread is holding a lock, you can call Tag.current_advisory_lock which will return the name of the current lock. If no lock is currently held, .current_advisory_lock returns nil.

Installation

Add this line to your application's Gemfile:

gem 'with_advisory_lock'

And then execute:

$ bundle

Lock Types

First off, know that there are lots of different kinds of locks available to you. Pick the finest-grain lock that ensures correctness. If you choose a lock that is too coarse, you are unnecessarily blocking other processes.

Advisory locks

These are named mutexes that are inherently "application level"—it is up to the application to acquire, run a critical code section, and release the advisory lock.

Row-level locks

Whether optimistic or pessimistic, row-level locks prevent concurrent modification to a given model.

If you're building a CRUD application, this will be your most commonly used lock.

Table-level locks

Provided through something like the monogamy gem, these prevent concurrent access to any instance of a model. Their coarseness means they aren't going to be commonly applicable, and they can be a source of deadlocks.

FAQ

Transactions and Advisory Locks

Advisory locks with MySQL and PostgreSQL ignore database transaction boundaries.

You will want to wrap your block within a transaction to ensure consistency.

MySQL doesn't support nesting

With MySQL (at least <= v5.5), if you ask for a different advisory lock within a with_advisory_lock block, you will be releasing the parent lock (!!!). A NestedAdvisoryLockErrorwill be raised in this case. If you ask for the same lock name, with_advisory_lock won't ask for the lock again, and the block given will be yielded to.

There are many lock-* files in my project directory after test runs

This is expected if you aren't using MySQL or Postgresql for your tests. See issue 3.

SQLite doesn't have advisory locks, so we resort to file locking, which will only work if the FLOCK_DIR is set consistently for all ruby processes.

In your spec_helper.rb or minitest_helper.rb, add a before and after block:

before do
  ENV['FLOCK_DIR'] = Dir.mktmpdir
end

after do
  FileUtils.remove_entry_secure ENV['FLOCK_DIR']
end

Changelog

3.0.0

• Added jruby/PostgreSQL support for Rails 4.x
• Reworked threaded tests to allow jruby tests to pass

API changes

• yield_with_lock_and_timeout and yield_with_lock now return instances of WithAdvisoryLock::Result, so blocks that return false are not misinterpreted as a failure to lock. As this changes the interface (albeit internal methods), the major version number was incremented.
• with_advisory_lock_result was introduced, which clarifies whether the lock was acquired versus the yielded block returned false.

2.0.0

• Lock timeouts of 0 now attempt the lock once, as per suggested by Jon Leighton and implemented by Abdelkader Boudih. Thanks to both of you!
• Pull request 11 fixed a downstream issue with jruby support! Thanks, Aaron Todd!
• Added Travis tests for jruby
• Dropped support for Rails 3.0, 3.1, and Ruby 1.8.7, as they are no longer receiving security patches. See http://rubyonrails.org/security/ for more information. This required the major version bump.
• Refactored advisory_lock_exists? to use existing functionality
• Fixed sqlite's implementation so parallel tests could be run against it

1.0.0

• Releasing 1.0.0. The interface will be stable.
• Added advisory_lock_exists?. Thanks, Sean Devine, for the great pull request!
• Added Travis test for Rails 4.1

0.0.10

• Explicitly added MIT licensing to the gemspec.

0.0.9

• Merged in Postgis Adapter Support to address issue 7 Thanks for the pull request, Abdelkader Boudih!
• The database switching code had to be duplicated by Closure Tree, so I extracted a new WithAdvisoryLock::DatabaseAdapterSupport one-trick pony.
• Builds were failing on Travis, so I introduced a global lock prefix that can be set with the WITH_ADVISORY_LOCK_PREFIX environment variable. I'm not going to advertise this feature yet. It's a secret. Only you and I know, now. shhh

0.0.8

• Addressed issue 5 by using a deterministic hash for Postgresql + MRI >= 1.9. Thanks for the pull request, Joel Turkel!
• Addressed issue 2 by using a cache-busting query for MySQL and Postgres to deal with AR value caching bug. Thanks for the pull request, Jaime Giraldo!
• Addressed issue 4 by adding support for em-postgresql-adapter. Thanks, lestercsp!

(Hey, github—your notifications are WAY too easy to ignore!)

0.0.7

• Added Travis tests for Rails 3.0, 3.1, 3.2, and 4.0
• Fixed MySQL bug with select_value returning a string instead of an integer when using AR 3.0.x

0.0.6

• Only require ActiveRecord >= 3.0.x
• Fixed MySQL error reporting

0.0.5

• Asking for the currently acquired advisory lock doesn't re-ask for the lock now.
• Introduced NestedAdvisoryLockError when asking for different, nested advisory locksMySQL

0.0.4

• Moved require into on_load, which should speed loading when AR doesn't have to spin up

0.0.3

• Fought with ActiveRecord 3.0.x and 3.1.x. You don't want them if you use threads—they fail predictably.

0.0.2

• Added warning log message for nested MySQL lock calls
• Randomized lock wait time, which can help ameliorate lock contention

0.0.1

• First whack